Share via

XML and ADO.NET Types in Data Contracts

  • Article

The Windows Communication Foundation (WCF) data contract model supports certain types that represent XML directly. When these types are serialized to XML, the serializer writes out the XML contents of these types without any further processing. Supported types are XmlElement, arrays of XmlNode (but not the XmlNode type itself), as well as types that implement IXmlSerializable. The DataSet and DataTable type, as well as typed datasets, are commonly used in database programming. These types implement the IXmlSerializable interface and are therefore serializable in the data contract model. Some special considerations for these types are listed at the end of this topic.

XML Types

Xml Element

The XmlElement type is serialized using its XML contents. For example, using the following type.

<DataContract([Namespace] := "http://schemas.contoso.com")>  _
Public Class MyDataContract
    <DataMember()>  _
    Public myDataMember As XmlElement
    
    Public Sub TestClass() 
        Dim xd As New XmlDocument()
        myDataMember = xd.CreateElement("myElement")
        myDataMember.InnerText = "myContents"
        myDataMember.SetAttribute("myAttribute", "myValue")
    
    End Sub 
End Class 

[DataContract(Namespace=@"http://schemas.contoso.com")]
public class MyDataContract
{
    [DataMember]
    public XmlElement myDataMember;
    public void TestClass()
    {
        XmlDocument xd = new XmlDocument();
        myDataMember = xd.CreateElement("myElement");
        myDataMember.InnerText = "myContents";
        myDataMember.SetAttribute
         ("myAttribute","myValue");
    }
}

This is serialized to XML as follows:

<MyDataContract xmlns="http://schemas.contoso.com">
    <myDataMember>
        <myElement  myAttribute="myValue">
            myContents
        </myElement>
    </myDataMember>
</MyDataContract>

Notice that a wrapper data member element <myDataMember> is still present. There is no way of removing this element in the data contract model. The serializers that handle this model (the DataContractSerializer and NetDataContractSerializer) may emit special attributes into this wrapper element. These attributes include the standard XML Schema Instance "nil" attribute (allowing the XmlElement to be null) and the "type" attribute (allowing XmlElement to be used polymorphically). Also, the following XML attributes are specific to WCF: "Id", "Ref", "Type" and "Assembly". These attributes may be emitted to support using the XmlElement with the object graph preservation mode enabled, or with the NetDataContractSerializer. (For more information about the object graph preservation mode, see Serialization and Deserialization.)

Arrays or collections of XmlElement are allowed and are handled as any other array or collection. That is, there is a wrapper element for the entire collection, and a separate wrapper element (similar to <myDataMember> in the preceding example) for each XmlElement in the array.

On deserialization, an XmlElement is created by the deserializer from the incoming XML. A valid parent XmlDocument is provided by the deserializer.

Make sure that the XML fragment that is deserialized to an XmlElement defines all prefixes that it uses and does not rely on any prefix definitions from ancestor elements. This is a concern only when using the DataContractSerializer to access XML from a different (non-DataContractSerializer) source.

When used with the DataContractSerializer, the XmlElement may be assigned polymorphically, but only to a data member of type Object. Even though it implements IEnumerable, an XmlElement cannot be used as a collection type and cannot be assigned to an IEnumerable data member. As with all polymorphic assignments, the DataContractSerializer emits the data contract name in the resulting XML – in this case, it is "XmlElement" in the "http://schemas.datacontract.org/2004/07/System.Xml" namespace.

With the NetDataContractSerializer, any valid polymorphic assignment of XmlElement (to Object or IEnumerable) is supported.

Do not attempt to use either of the serializers with types derived from XmlElement, whether they are assigned polymorphically or not.

Array of XmlNode

Using arrays of XmlNode is very similar to using XmlElement. Using arrays of XmlNode gives you more flexibility than using XmlElement. You can write multiple elements inside the data member wrapping element. You can also inject content other than elements inside of the data member wrapping element, such as XML comments. Finally, you can put attributes into the wrapping data member element. All this can be achieved by populating the array of XmlNode with specific derived classes of XmlNode such as XmlAttribute, XmlElement or XmlComment. For example, using the following type.

<DataContract([Namespace] := "http://schemas.contoso.com")>  _
Public Class MyDataContract
    <DataMember()>  _
    Public myDataMember(3) As XmlNode
    
    Public Sub TestClass() 
        Dim xd As New XmlDocument()
        Dim xe As XmlElement = xd.CreateElement("myElement")
        xe.InnerText = "myContents"
        xe.SetAttribute("myAttribute", "myValue")
        
        Dim atr As XmlAttribute = xe.Attributes(0)
        Dim cmnt As XmlComment = xd.CreateComment("myComment")
        
        myDataMember(0) = atr
        myDataMember(1) = cmnt
        myDataMember(2) = xe
        myDataMember(3) = xe
    
    End Sub 
    
End Class 

[DataContract(Namespace="http://schemas.contoso.com")]
public class MyDataContract
{
    [DataMember]
    public XmlNode[] myDataMember = new XmlNode[4];
    public void TestClass()
    {
        XmlDocument xd = new XmlDocument();
        XmlElement xe = xd.CreateElement("myElement");
        xe.InnerText = "myContents";
        xe.SetAttribute
         ("myAttribute","myValue");
      
        XmlAttribute atr = xe.Attributes[0];
        XmlComment cmnt = xd.CreateComment("myComment");
        
      myDataMember[0] = atr;
      myDataMember[1] = cmnt;
      myDataMember[2] = xe;
      myDataMember[3] = xe;
    }
}

When serialized, the resulting XML is similar to the following code.

<MyDataContract xmlns="http://schemas.contoso.com">
  <myDataMember myAttribute="myValue">
     <!--myComment-->
     <myElement  myAttribute="myValue">
 myContents
     </myElement>
     <myElement  myAttribute="myValue">
       myContents
     </myElement>
  </myDataMember>
</MyDataContract>

Note that the data member wrapper element <myDataMember> contains an attribute, a comment, and two elements. These are the four XmlNode instances that were serialized.

An array of XmlNode that results in invalid XML cannot be serialized. For example, an array of two XmlNode instances where the first one is an XmlElement and the second one is an XmlAttribute is invalid, because this sequence does not correspond to any valid XML instance (there is no place to attach the attribute to).

On deserialization of an array of XmlNode, nodes are created and populated with information from the incoming XML. A valid parent XmlDocument is provided by the deserializer. All nodes are deserialized, including any attributes on the wrapper data member element, but excluding the attributes placed there by the WCF serializers (such as the attributes used to indicate polymorphic assignment). The caveat about defining all namespace prefixes in the XML fragment applies to the deserialization of arrays of XmlNode just like it does to deserializing XmlElement.

When using the serializers with object graph preservation turned on, object equality is only preserved on the level of XmlNode arrays, not individual XmlNode instances.

Do not attempt to serialize an array of XmlNode where one or more of the nodes is set to null. It is permitted for the entire array member to be null, but not for any individual XmlNode contained in the array. If the entire array member is null, the wrapper data member element contains a special attribute that indicates that it is null. On deserialization, the entire array member also becomes null.

Only regular arrays of XmlNode are treated specially by the serializer. Data members declared as other collection types that contain XmlNode, or data members declared as arrays of types derived from XmlNode, are not treated specially. Thus, they are normally not serializable unless they also meet one of the other criteria for serializing.

Arrays or collections of arrays of XmlNode are allowed. There is a wrapper element for the entire collection, and a separate wrapper element (similar to <myDataMember> in the preceding example) for each array of XmlNode in the outer array or collection.

Populating a data member of type Array of Object or Array of IEnumerable with XmlNode instances does not result in the data member being treated as an Array of XmlNode instances. Each array member is serialized separately.

When used with the DataContractSerializer, arrays of XmlNode can be assigned polymorphically, but only to a data member of type Object. Even though it implements IEnumerable, an array of XmlNode cannot be used as a collection type and be assigned to an IEnumerable data member. As with all polymorphic assignments, the DataContractSerializer emits the data contract name in the resulting XML – in this case, it is "ArrayOfXmlNode" in the "http://schemas.datacontract.org/2004/07/System.Xml" namespace. When used with the NetDataContractSerializer, any valid assignment of an XmlNode array is supported.

Schema Considerations

For details about the schema mapping of XML types, see Data Contract Schema Reference. This section provides a summary of the important points.

A data member of type XmlElement is mapped to an element defined using the following anonymous type.

<xsd:complexType>
   <xsd:sequence>
      <xsd:any minOccurs="0" processContents="lax" />
   </xsd:sequence>
</xsd:complexType>

A data member of type Array of XmlNode is mapped to an element defined using the following anonymous type.

<xsd:complexType mixed="true">
   <xsd:sequence>
      <xsd:any minOccurs="0" maxOccurs="unbounded" processContents="lax" />
   </xsd:sequence>
   <xsd:anyAttribute/>
</xsd:complexType>

Types Implementing the IXmlSerializable Interface

Types that implement the IXmlSerializable interface are fully supported by the DataContractSerializer. The XmlSchemaProviderAttribute attribute should always be applied to these types to control their schema.

There are three varieties of types that implement IXmlSerializable: types that represent arbitrary content, types that represent a single element, and legacy DataSet types.

  • Content types use a schema provider method specified by the XmlSchemaProviderAttribute attribute. The method does not return null, and the IsAny property on the attribute is left at its default value of false. This is the most common usage of IXmlSerializable types.

  • Element types are used when an IXmlSerializable type must control its own root element name. To mark a type as an element type, either set the IsAny property on the XmlSchemaProviderAttribute attribute to true or return null from the schema provider method. Having a schema provider method is optional for element types – you may specify null instead of the method name in the XmlSchemaProviderAttribute. However, if IsAny is true and a schema provider method is specified, the method must return null.

  • Legacy DataSet types are IXmlSerializable types that are not marked with the XmlSchemaProviderAttribute attribute. Instead, they rely on the GetSchema method for schema generation. This pattern is used for the DataSet type and its typed dataset derives a class in earlier versions of the .NET Framework, but is now obsolete and is supported only for legacy reasons. Do not rely on this pattern and always apply the XmlSchemaProviderAttribute to your IXmlSerializable types.

IXmlSerializable Content Types

When serializing a data member of a type that implements IXmlSerializable and is a content type as defined previously, the serializer writes the wrapper element for the data member and pass control to the WriteXml method. The WriteXml implementation can write any XML, including adding attributes to the wrapper element. After WriteXml is done, the serializer closes the element.

When deserializing a data member of a type that implements IXmlSerializable and is a content type as defined previously, the deserializer positions the XML reader on the wrapper element for the data member and pass control to the ReadXml method. The method must read the entire element, including the start and end tags. Make sure your ReadXml code handles the case where the element is empty. Additionally, your ReadXml implementation should not rely on the wrapper element being named a particular way. The name is chosen by the serializer can vary.

It is permitted to assign IXmlSerializable content types polymorphically, for example, to data members of type Object. It is also permitted for the type instances to be null. Finally, it is possible to use IXmlSerializable types with object graph preservation enabled and with the NetDataContractSerializer. All these features require the WCF serializer to attach certain attributes into the wrapper element ("nil" and "type" in the XML Schema Instance namespace and "Id", "Ref", "Type" and "Assembly" in a WCF-specific namespace).

Attributes to Ignore when Implementing ReadXml

Before passing control to your ReadXml code, the deserializer examines the XML element, detects these special XML attributes, and acts on them. For example, if "nil" is true, a null value is deserialized and ReadXml is not called. If polymorphism is detected, the contents of the element are deserialized as if it was a different type. The polymorphically assigned type’s implementation of ReadXml is called. In any case, a ReadXml implementation should ignore these special attributes because they are handled by the deserializer.

Schema Considerations for IXmlSerializable Content Types

When exporting schema an IXmlSerializable content type, the schema provider method is called. An XmlSchemaSet is passed to the schema provider method. The method can add any valid schema to the schema set. The schema set contains the schema that is already known at the time when schema export occurs. When the schema provider method must add an item to the schema set, it must determine if an XmlSchema with the appropriate namespace already exists in the set. If it does, the schema provider method must add the new item to the existing XmlSchema. Otherwise, it must create a new XmlSchema instance. This is important if arrays of IXmlSerializable types are being used. For example, if you have an IXmlSerializable type that gets exported as type "A" in namespace "B", it is possible that by the time the schema provider method is called the schema set already contains the schema for "B" to hold the "ArrayOfA" type.

In addition to adding types to the XmlSchemaSet, the schema provider method for content types must return a non-null value. It can return an XmlQualifiedName that specifies the name of the schema type to use for the given IXmlSerializable type. This qualified name also serves as the data contract name and namespace for the type. It is permitted to return a type that does not exist in the schema set immediately when the schema provider method returns. However, it is expected that by the time all related types are exported (the Export method is called for all relevant types on the XsdDataContractExporter and the Schemas property is accessed), the type exists in the schema set. Accessing the Schemas property before all relevant Export calls have been made can result in an XmlSchemaException. For more information about the export process, see Exporting Schemas from Classes.

The schema provider method can also return the XmlSchemaType to use. The type may or may not be anonymous. If it is anonymous, the schema for the IXmlSerializable type is exported as an anonymous type every time the IXmlSerializable type is used as a data member. The IXmlSerializable type still has a data contract name and namespace. (This is determined as described in Data Contract Names except that the DataContractAttribute attribute cannot be used to customize the name.) If it is not anonymous, it must be one of the types in the XmlSchemaSet. This case is equivalent to returning the XmlQualifiedName of the type.

Additionally, a global element declaration is exported for the type. If the type does not have the XmlRootAttribute attribute applied to it, the element has the same name and namespace as the data contract, and its "nillable" property is true. The only exception to this is the schema namespace ("http://www.w3.org/2001/XMLSchema") – if the type’s data contract is in this namespace, the corresponding global element is in the blank namespace because it is forbidden to add new elements to the schema namespace. If the type has the XmlRootAttribute attribute applied to it, the global element declaration is exported using the following: ElementName, Namespace and IsNullable properties. The defaults with XmlRootAttribute applied are the data contract name, a blank namespace and "nillable" being true.

The same global element declaration rules apply to legacy dataset types. Note that the XmlRootAttribute cannot override global element declarations added through custom code, either added to the XmlSchemaSet using the schema provider method or through GetSchema for legacy dataset types.

IXmlSerializable Element Types

IXmlSerializable element types have either the IsAny property set to true or have their schema provider method return null.

Serializing and deserializing an element type is very similar to serializing and deserializing a content type. However, there are some important differences:

  • The WriteXml implementation is expected to write exactly one element (which could of course contain multiple child elements). It should not be writing attributes outside of this single element, multiple sibling elements or mixed content. The element may be empty.

  • The ReadXml implementation should not read the wrapper element. It is expected to read the one element that WriteXml produces.

  • When serializing an element type regularly (for example, as a data member in a data contract), the serializer outputs a wrapper element before calling WriteXml, as with content types. However, when serializing an element type at the top level, the serializer does not normally output a wrapper element at all around the element that WriteXml writes, unless a root name and namespace were explicitly specified when constructing the serializer in the DataContractSerializer or NetDataContractSerializer constructors. For more information, see Serialization and Deserialization.

  • When serializing an element type at the top level without specifying the root name and namespace at construction time, WriteStartObject and WriteEndObject essentially does nothing and WriteObjectContent calls WriteXml. In this mode, the object being serialized cannot be null and cannot be polymorphically assigned. Also, object graph preservation cannot enabled and the NetDataContractSerializer cannot be used.

  • When deserializing an element type at the top level without specifying the root name and namespace at construction time, IsStartObject returns true if it can find the start of any element. ReadObject with the verifyObjectName parameter set to true behaves in the same way as IsStartObject before actually reading the object. ReadObject then passes control to ReadXml method.

The schema exported for element types is the same as for the XmlElement type as described in an earlier section, except that the schema provider method can add any additional schema to the XmlSchemaSet as with content types. Using the XmlRootAttribute attribute with element types is not allowed, and global element declarations are never emitted for these types.

Differences from the XmlSerializer

The IXmlSerializable interface and the XmlSchemaProviderAttribute and XmlRootAttribute attributes are also understood by the XmlSerializer . However, there are some differences in how these are treated in the data contract model. The important differences are summarized in the following:

  • The schema provider method must be public to be usable in the XmlSerializer, but does not have to be public to be usable in the data contract model.

  • The schema provider method is called when IsAny is true in the data contract model but not with the XmlSerializer.

  • When the XmlRootAttribute attribute is not present for content or legacy dataset types, the XmlSerializer exports a global element declaration in the blank namespace. In the data contract model, the namespace used is normally the data contract namespace as described earlier.

Be aware of these differences when creating types that are used with both serialization technologies.

Importing IXmlSerializable Schema

When importing a schema generated from IXmlSerializable types, there are a few possibilities:

  • The generated schema may be a valid data contract schema as described in Data Contract Schema Reference. In this case, schema can be imported as usual and regular data contract types are generated.

  • The generated schema may not be a valid data contract schema. For example, your schema provider method may generate schema that involves XML attributes which are not supported in the data contract model. In this case, you can import the schema as IXmlSerializable types. This import mode is not on by default but can easily be enabled – for example, with the /importXmlTypes command-line switch to the ServiceModel Metadata Utility Tool (Svcutil.exe). This is described in detail in the Importing Schema to Generate Classes. Note that you must work directly with the XML for your type instances. You may also consider using a different serialization technology that supports a wider range of schema – see the topic on using the XmlSerializer.

  • You may want to reuse your existing IXmlSerializable types in the proxy instead of generating new ones. In this case, the referenced types feature described in the Importing Schema to Generate Types topic can be used to indicate the type to reuse. This corresponds to using the /reference switch on svcutil.exe, which specifies the assembly that contains the types to reuse.

Representing Arbitrary XML in Data Contracts

The XmlElement, Array of XmlNode and IXmlSerializable types allow you to inject arbitrary XML into the data contract model. The DataContractSerializer and NetDataContractSerializer pass this XML content on to the XML writer in use, without interfering in the process. However, the XML writers may enforce certain restrictions on the XML that they write. Specifically, here are some important examples:

  • The XML writers do not typically allow an XML document declaration (for example, <?xml version=’1.0’ ?>) in the middle of writing another document. You cannot take a full XML document and serialize it as an Array of XmlNode data member. To do this, you have to either strip out the document declaration or use your own encoding scheme to represent it.

  • All of the XML writers supplied with WCF reject XML processing instructions (<? … ?>) and document type definitions (<! … >), because they are not allowed in SOAP messages. Again, you can use your own encoding mechanism to get around this restriction. If you must include these in your resultant XML, you can write a custom encoder that uses XML writers that support them.

  • When implementing WriteXml, avoid calling WriteRaw method on the XML writer. WCF uses a variety of XML encodings (including binary), it is very difficult or impossible to use WriteRaw such that the result is usable in any encoding.

  • When implementing WriteXml, avoid using the WriteEntityRef and WriteNmToken methods that are unsupported on the XML writers supplied with WCF.

Using DataSet, Typed DataSet and DataTable

Using these types is fully supported in the data contract model. When using these types, consider the following points:

  • The schema for these types (especially DataSet and its typed derived classes) may not be interoperable with some non-WCF platforms, or may result in poor usability when used with these platforms. Additionally, using the DataSet type may have performance implications. Finally, it may make it more difficult for you to version your application in the future. Consider using explicitly defined data contract types instead of DataSet types in your contracts.

  • When importing DataSet or DataTable schema, it is important to reference these types. With the Svcutil.exe command-line tool, this can be accomplished by passing the System.Data.dll assembly name to the /reference switch. If importing typed dataset schema, you must reference the typed dataset’s type. With Svcutil.exe, pass the location of the typed dataset’s assembly to the /reference switch. For more information about referencing types, see the Importing Schema to Generate Classes.

Support for typed DataSets in the data contract model is limited. Typed DataSets can be serialized and deserialized and can export their schema. However, the Data Contract schema import is unable to generate new typed DataSet types from the schema, as it can only reuse existing ones. You can point to an existing typed DataSet by using the /r switch on Svcutil.exe. If you attempt to use a Svcutil.exe without the /r switch on a service that uses a typed dataset, an alternative serializer (XmlSerializer) is automatically selected. If you must use the DataContractSerializer and must generate DataSets from schema, you can use the following procedure: generate the typed DataSet types (by using the Xsd.exe tool with the /d switch on the service), compile the types, and then point to them using the /r switch on Svcutil.exe.

See Also

Reference

DataContractSerializer
IXmlSerializable

Concepts

Using Data Contracts
Types Supported by the Data Contract Serializer